Rakuten's open RAN diversity claims ring hollow
The UK town of Weybridge is known as a place of high-speed attractions, with a rich history of car racing and aircraft design. But the latest arrival that courts a link with speed and dynamism comes from telecom and tech, not motorsports and aviation. Rakuten Symphony, a division of the well-known Japanese ecommerce player, is setting itself up as an alternative to 147-year-old Ericsson and 158-year-old Nokia for the network products that support billions of 4G and 5G connections worldwide. It's a market Rakuten executives believe is ripe for disruption.
They have an ally in the UK government, present at Rakuten's offices in Weybridge this week for the opening of a new lab, where partners can try out products and see if they fit. Today, telcos usually buy a ready-made radio access network (RAN) for any given site from a single big vendor – be it Ericsson, Nokia or China's Huawei. Rakuten has swung behind the concept of open RAN, a Lego-like way of combining different suppliers. The UK's Tory government is a fan.
Enthusiasm stems from the decision in 2020 to restrict and ultimately ban Huawei from the UK on grounds of national security. In a market for critical infrastructure dominated by three big players, that effectively created a Nordic duopoly. To policy wonks, open RAN promises diversity.
"The reason we think open RAN is an exciting means to deliver that is because it ultimately removes the barriers to entry that we see as contributing to the consolidation we've seen in the market," said Scott Bailey, whose job title – head of telecom diversification for the department of science, innovation and technology – highlights the main focus for authorities. "Having more vendors in the supply chain helps to increase resilience."
But the whole premise is deeply flawed. The open interfaces championed by Rakuten aid specialization, not diversity, even if they boost the overall number of market players. The previous absence of those interfaces forced telcos to rely on vendors that could provide everything. Open RAN, essentially, allows a supplier focused on one part of the RAN to compete for contracts, knowing other specialists can handle what it can't.
It provides no guarantee there will be a multiplicity of suppliers in each separate domain, and some of these domains already look uncompetitive. The standout example relates to virtualization, where software is deployed on general-purpose processors (GPPs) after it has been decoupled from customized silicon. Benefits include being able to host RAN software on the same cloud platform as other network and IT workloads, say advocates. But Intel controls more than 70% of the market for GPPs used in data centers, while AMD, an Intel licensee, serves most of the rest. What's more, Intel recently boasted a 99% share of the nascent virtual RAN market.
Rabih Dabboussi, the chief business officer of Rakuten Symphony, concedes that is a problem while insisting it is an improvement on the status quo. "It's a good point you bring, but again this is not a closed system," he told Light Reading during a session with analysts and reporters at the lab opening. "At least it's more diversified than a proprietary ASIC [application-specific integrated circuit] developed to serve vendor-specific architecture."
"The second part is that general purpose is commoditized from an economic and supply-chain perspective," he continued. "Because of the volume used, there is a massive amount of R&D invested behind it to provide the next and the next and the next generation. That is why we've seen these CPUs [central processing units] pushing the limits of performance of an ASIC developed by a specific vendor."
Both points are controversial. Even if it happens, an overtaking of ASICs by GPPs on performance would not aid diversity and could even have the opposite effect. Dabboussi may be right to argue that a GPP is not vendor-specific – in theory, it should be able to host any company's RAN software – but Intel remains the dominant provider of a key component. Unless this changes, problems at Intel in a world of virtual RAN ubiquity would mean problems for the whole system. That hardly sounds like the "resilience" Bailey seeks.
The risks were illustrated when delays at Intel on the development of ten-nanometer chips (the smaller, the more advanced) upended Nokia's 5G business back in 2019. The chips in question were ASICs intended for Nokia's 5G basestations and the hold-up forced the company to fall back on more expensive field programmable gate arrays (FPGAs) designed by Xilinx, now a part of AMD. Margins were squeezed, market share fell and Rajeev Suri, Nokia's CEO at the time, ultimately quit.
His big mistake was to have relied solely on Intel. "Now what has changed is that we no longer work with only one supplier but with two other SoC [system-on-a-chip] partners making custom silicon for us in mobile networks," he explained on a call with analysts in February 2020 – months before he left – after Nokia had introduced Broadcom and Marvell alongside Intel.
The performance claims about GPPs are spurious because even Intel has recognized the need for "accelerators" to support compute-intensive RAN software normally sold with ASICs. These accelerators would introduce other silicon for that RAN software to overcome the drawbacks of relying solely on GPPs. But Intel's preferred options – named "lookaside" and "integrated" (critics say the latter is just a rebadged lookaside) – are firmly anchored to the GPP. By offloading one layer of the RAN software onto custom silicon, an alternative called "inline" would have no need for the GPP whatsoever in that part of the network.
For this reason, lookaside and integrated worry some operators on competition grounds alone. "If you stay with lookaside, you may be creating lock-in to Intel as a supplier of silicon," said Yago Tenorio, Vodafone's network architecture director, during a conversation with Light Reading at this year's Mobile World Congress (MWC). Yet Rakuten has been firmly in Intel's camp.
This has also meant building its RAN software, based on the Altiostar business that Rakuten fully acquired in 2021, on top of FlexRAN, Intel's reference design. Tenorio regards it as an egregious example of lock-in. "You have to design your software in a particular way using the instruction set FlexRAN gives you," he told Light Reading last year. "Once coded for FlexRAN, it is not portable for another accelerator – you can only use Intel from that point on."
Good consolidation, bad consolidation
Regardless of the nitty gritty, the push for RAN diversity is at odds with not only historical trends but also the desire for consolidation in Europe's telecom sector, as noted by Francis Haysom, an analyst with Appledore Research, during a panel session at Rakuten's lab opening.
The response of ETNO, a lobby group for big European operators, seems to be that some forms of consolidation are more acceptable than others. "I think consolidation in the telecom sector among telecom operators is not bad because right now there is a lot of fragmentation and a lot of smaller players that lack the financial capability to go out there and innovate," said Alessandro Gropelli, ETNO's deputy director-general, when challenged by Haysom on the point about European consolidation.
"We know that competition can stimulate innovation, can stimulate investment, but then there is a breaking point and if there is too much it can weaken innovation and investment," he continued. The obvious retort is to ask why a three-player market is fine for mobile consumers when that number of RAN suppliers to operators is clearly deemed insufficient. Excluding China's Huawei and ZTE, the world has at least three companies that can build 4G and 5G networks in Ericsson, Nokia and Samsung.
Dabboussi and Gropelli are right about economies of scale and the risks posed by a fragmented market to innovation and investment. Low-priced mobile services have taken off partly because these huge global companies have been able to drive down unit costs while injecting multi-billion-dollar sums into research and development each year.
The problem for smaller companies was noted at an event last October where Mavenir, an open RAN vendor, complained to Marvell, a semiconductor maker, that it paid a high price per chip compared with its bigger rivals. "Volume matters anywhere in the industry, and we are talking about millions of units," responded Joel Brand, Marvell's senior director of product marketing. "When Mavenir becomes the same size as Nokia, they will get the same price."
Economies of scale have resulted from a drive for global standards and years of consolidation that left the industry with just a handful of global suppliers. What no one in the open RAN community has been able to explain is how tomorrow's RAN market will be able to support considerable diversity when yesterday's industry could not – unless governments intervene to prevent consolidation from happening.
The warning signs are already there. Former specialists, Mavenir included, are now pitching themselves as "end-to-end" suppliers. The next logical step is the takeover of one such end-to-end supplier by another to realize "synergies." This seems even likelier if operators seek to build open RAN networks while avoiding the complications of dealing with multiple vendors.
"I think it's good to have greater choice but that doesn't necessarily mean we'll have a van with six different vendors' worth of spares in it," said Howard Watson, the chief security and networks officer of BT, during a conversation at MWC. "There is an operational inefficiency that comes from that."
None of this means Rakuten Symphony is doomed to fail. It has a charismatic and capable CEO in Tareq Amin and it belongs to a large group that made about $14.6 billion in revenues last year (although it has also racked up $5 billion worth of losses in four years on its own mobile rollout in Japan).
After landing a few important deals, it reported sales of $476 million last year, nearly seven times what it made in 2021. Talks with big European operators are going well, said Nastasi Karaiskos, the managing director of Rakuten Symphony's UK business. But Rakuten's own target of capturing 25% of the RAN market belies all the talk of diversity – perhaps the most-used word at the lab opening. It is time that was dropped.
- The virtual RAN benefits are as clear as mud, unless you're Intel
- Good luck building a virtual, open RAN – there's no such thing
- Vodafone slams Intel and its chip rivals on standardization
- Mavenir unhappy about chip prices for smaller open RAN players
- Rakuten's Amin targets 25% of RAN market in bid to unseat giants
— Iain Morris, International Editor, Light Reading